prompt-engineer

To write cross-provider system prompts, use XML-like sections to separate role, constraints, examples, and output format. This portable structure ensures your AI agent instructions maintain consistency and performance across Claude, GPT, and Gemini APIs.

The best way to structure tool descriptions for LLM APIs is to use standardized patterns and templates that separate function definitions from contextual guidance. This improves tool-use accuracy and maintainability across different AI providers.

Prompt engineering for AI agents does not require entirely different approaches for each provider. By applying context engineering and subagent design best practices, you can create reliable prompts that generalize across leading LLM APIs without becoming brittle.

To test prompts for reliability across multiple LLM providers, start by editing a minimal system prompt and a simple tool description using the XML structure, then validate consistency by running the same prompts against Claude, GPT, and Gemini APIs.

AI agent prompts fail to generalize across different LLM APIs when they lack a portable, structured approach. Without using XML-like sections to separate components and applying context engineering, prompts become brittle and perform inconsistently across providers.